What breaks presence is often inconsistency in the details
Aliases: BiP · presence failure · local inconsistency · out-of-body moment
What it is
In a room with enough polygons, presence usually dies on a menu with no contact shadow, a hand through a tabletop, wallpaper that slides a little on a lateral move, a chair scrape in the lab — not on a global verdict of “not real enough.” Slater called these events breaks in presence (BiP): the moment a person realises they are looking at a display apparatus rather than occupying an environment. What breaks it is often a local contradiction inside an already-adopted world, not one missing tier of global fidelity.
A break is a moment, not a score. A session can contain many brief breaks and recoveries; an end-of-session questionnaire averages them away and makes the experience look flat.
Why it happens
Perception collapses current input into a hypothesis of one coherent world. Once adopted, the hypothesis is cheap to maintain — until evidence arrives that cannot coexist with it. A hand through a solid, sound from the wrong direction, interface elements that ignore occlusion and gravity, the world lagging a half-beat on a head turn, are that kind of evidence. They cut deeper than “not enough polygons” because they happen inside the hypothesis: the person is already explaining input as “this is a room,” the local contradiction falsifies the explanation, and the system switches to “this is a screen on my face.”
When global fidelity is too low, the hypothesis may never form; that is a different failure. After it has formed, the damage comes from inconsistency: a lamp with no reflection, an open door with no corridor sound, a line of system type floating in mid-air. Detail gets its weight from being treated as a test of an already-adopted model, not from its share of the pixel budget.
Studying it
The classic method lets people press a button during the experience to report “I was not there just then,” then aligns presses with logs and codes break type and time. Slater and Steed’s BiP counts are this path. The complement is a critical-incident interview: no checklist, “at which moment did you remember you were in the lab,” then hunt the matching visual or physical contradiction in the scene.
Independent variables: kind of local consistency error (occlusion, physics, spatial UI, external sound), amplitude and duration of the error. Dependent variables: BiP count and spacing, time from break to a further “I am there” report, end-of-session questionnaires (as a contrast, not as the sole measure).
Do not compare “with versus without flaws” on a total score alone. A two-second interpenetration can wreck the next minute of report and leave only noise on the average. Eye movements toward the contradiction help confirm that the break was seen, not only rationalised afterwards.
Where it stops holding
When the hypothesis has not yet formed, detail inconsistency has nothing to break — loading meshes, an explicit debug camera, someone still setting IPD. Some breaks are lab ecology (the experimenter talking, a fan, floor vibration) and vanish at home; lab BiP rates cannot be written down as content defects. Conversely, a notification banner at home or a family member opening the door is a break source the lab will not measure. Developers who know the engine recode interpenetration as “an engine problem” rather than “I am not there,” and their BiP reports run low. Deliberate stylisation (cartoon physics, UI as magic) can avoid breaks if it is internally consistent; detail becomes a crack only when one rule in the same experience slaps another.
Applying it
- Treat consistency errors as the defect list, not image quality: interpenetration, missing contact shadows, system type that is not spatially located, sound that does not match action. Close these before raising texture resolution.
- Ban world-foreign system layers (account, update, calibration) from the experience segment. If they must appear, give them an in-scene carrier; do not park a depthless line of type across the room.
- In playtests, log “the moment I fell out,” timestamped to a concrete object or sound, not to “it still wasn’t real enough overall.”
- How to check: put the headset on someone who did not build the project and ask only “call out every second you remembered the lab.” Mark the events in the scene. Any class of detail that appears more than twice gets fixed first — not a global quality bump.